OpenAI launched Prism, an AI-enhanced scientific workspace, on Tuesday. The free web app, accessible to anyone with a ChatGPT account, serves as a word processor and research tool for scientific papers, integrated with GPT-5.2 to assess claims, revise prose, and search prior research.
Prism requires human guidance and does not conduct research autonomously. OpenAI executives state that the tool accelerates work performed by human scientists. They compare Prism to coding interfaces such as Cursor and Windsurf, which assist developers in similar ways.
Kevin Weil, OpenAI’s VP for Science, spoke during a press call announcing the tool. He said, “I think 2026 will be for AI and science what 2025 was for AI and software engineering.” This statement reflects OpenAI’s perspective on the evolving role of AI in scientific workflows.
The launch coincides with increased scientific activity on OpenAI’s consumer products. ChatGPT receives an average of 8.4 million messages per week on advanced topics in the hard sciences. The company notes that determining the exact number from professional researchers proves challenging.
AI-assisted research gains prevalence among academic researchers. In mathematics, AI models have proven several long-standing Erdős problems. These proofs combine literature review with new applications of existing techniques. Debates persist regarding the significance of these proofs. Nonetheless, the results mark an early achievement for advocates of AI models and formal verification systems.
A statistics paper published in December employed GPT-5.2 Pro to establish new proofs for a central axiom of statistical theory. Human researchers handled prompting and verification of the model’s output. This approach limited their role to oversight.
OpenAI addressed this statistics paper in a blog post. The company praised the outcome as a model for human-AI collaboration in research. The post states, “In domains with axiomatic theoretical foundations, frontier models can help explore proofs, test hypotheses, and identify connections that might otherwise take substantial human effort to uncover.” OpenAI positions this as indicative of collaborative potential in structured fields.
